Refactor to remove circular import linked to build_message_definition

[Pierre-Sassoulas]

Work in progress for removing circular import with build_message_definition. Following the refactor to use checker inside the documentation function, MyPy suddenly raise a "MessagesHandlerMixIn" has no attribute "get_checkers"; that it was'nt raising before. MessagesHandlerMixIn never had a get_checkers function, but MyPy did not detect it before. the commit was just separating information retrieval from string creation in a MessagesHandlerMixIn function.

Description

• build_message_definition is used in pylint.checker.base_checker and also in message.MessageHandlerMixIn.print_full_documentation.
• The idea is to put this function in the checker packages, to break the cycle
• So we need to use a Checker instance in print_full_documentation
• Doing this refactoring makes MyPy realize that MessageHandlerMixIn do not have a get_checkers() function
  So there is probably a lot of thing to change to fix this.

Type of Changes

| ✓ | 🔨 Refactoring |

Related Issue

Based on #2809 but this is not mergeable yet so creating another PR.

[PCManticore]

Hey @Pierre-Sassoulas Sorry for the delayed review! This looks great, but there are some minor changes we need before getting this in, the major one being not using variable type assignments given that we still support 3.4 and 3.5.

[Pierre-Sassoulas]

Hi, @PCManticore thank you for the review. I took what you said into account but I'm not pushing the change yet because the BaseChecker.__doc__ function is not used/working as of now. What is done in MessageHandlerMixIn for printing documentation is really hard to understand and refactor. The printing and string creation was mixed and there are inputs informations that are different when it's called from doc/exts/pylint_extensions.py , than directly from MessageHandlerMixIn.get_full_documentation() so the information contained in checker are not sufficient for all cases. I'll ping you once the tests are passing.

[coveralls]

Coverage decreased (-0.2%) to 89.868% when pulling 3d1de86 on Pierre-Sassoulas:build-message-definition-in-checker into e3d4e80 on PyCQA:master.

[coveralls]

Coverage decreased (-0.06%) to 89.868% when pulling 0918e02 on Pierre-Sassoulas:build-message-definition-in-checker into ee228dd on PyCQA:master.

[Pierre-Sassoulas]

Hi @PCManticore. I took some time to fix this, there was a lot of disentanglement of print and output producing with implicit line feed when you use print to take into account. Sadly I could not keep the __doc__ function in BaseChecker, because if we were using a docstring in a checker it was overriding the BaseChecker.__doc__ function. I added a full exact output test during the development so we should produce exactly the same output after this refactor. We can't keep this test because the output is very long and change when checkers change. I could add some more string in the regex test though. I think there is a pre-existing problem with python 3.8-dev not caused by my change. I don't understand the problem with appveyor for python 3.4. Maybe it's preexisting too?

[PCManticore]

Hey @Pierre-Sassoulas This looks great, thank you for the hard work! The Python 3.4 failure is preexisting, as well as the 3.8 one. But I noticed that docs and mypy steps failed for you, can you fix those before getting this in?

[Pierre-Sassoulas]

Sure. I'll also rebase and fix the merging conflicts that appeared.

[Pierre-Sassoulas]

@PCManticore I fixed the merge conflict, mypy, and docs. I have no idea about the error by appveyor though, is it a completely different test than tox? How can I launch it locally?

[PCManticore]

Thanks for fixing this! Don't worry about appveyor, it's broken in master as well. In fact it is a form of TravisCI but for Windows, it's not possible to test it locally unless you have a Windows VM or a Windows OS.